Spot stuck card requests

# Aging of pending card requests

## What this skill covers — and what it doesn't

`get_requests` and `spendesk_analyze_requests` only return **card requests**: single-purchase card, subscription, multi-use card and its top-ups, physical card load.

**Not** covered, with no marker and no error: purchase order requests, travel budgets, expense claims, mileage allowances and invoices. Never present this report as "all purchase requests". For invoices waiting for approval, use the dedicated skill (`get_invoices`).

## Step 1 — Pending volume and amount

```json
{
  "tool": "spendesk_analyze_requests",
  "params": { "fromDate": "YYYY-MM-DD", "toDate": "YYYY-MM-DD", "groupBy": ["requestState", "month"] }
}
```

State the exact dates. The period applies to the requests' **creation date** and cannot exceed 366 days: take the last 12 months, and say so, because an older request still pending will not appear.

The `pending` row gives the pending count and amount; its monthly breakdown shows how old those requests are.

## Step 2 — List pending requests, with their approver

```json
{ "tool": "get_requests", "params": { "state": "pending", "withApproval": true, "pageSize": 100 } }
```

- The filter is called `state` (not `status`). `withApproval: true` adds the approval chain.
- A request's age = today − `createdAt`.
- Expected approver = the `approvers` of the step whose `state` is `current`.
- Amount: read `amountInFunctionalCurrency.formatted` to compare or total; `requestedAmount` is in the request's own currency. Both can be `null` (a request with no amount).
- Page with `page` while `hasNextPage` is `true`.

## Step 3 — Sort by age

| Band | Age |
|------|-----|
| 🔴 Critical | more than 30 days |
| 🟠 Urgent | 15 to 30 days |
| 🟡 To watch | 7 to 14 days |
| 🟢 Recent | less than 7 days |

For each band: number of requests and total amount in functional currency.

## Step 4 — Find where it is stuck

Group the pending requests:

- **by expected approver** — the most useful grouping: the same name on all the old requests points to the bottleneck;
- by requester (`requesterId`);
- by cost center (`costCentreId`, often `null`: say so rather than showing an empty group).

Resolve every ID before showing anything:

```json
{ "tool": "get_users", "params": { "ids": ["<userId>", "<requesterId>"] } }
```

`get_cost_centers` resolves cost centers, `get_supplier_by_id` resolves suppliers. Never show a bare ID; if one does not resolve, say so.

## Step 5 — Report

```
Pending card requests — <today's date> (created since <date>)
Total: N requests — X EUR

| Band | Requests | Amount |
| Request | Requester | Amount | Age | Expected approver | Link |

Chase first: requests older than 30 days, by approver.
```

`appUrl` is the direct link to each request: give it as returned.

## Limits to state in the report

- **No processing time.** A request carries no approval date and no last-updated date: you can measure how long a request has been waiting, not how long approved requests took. Do not compute an "average approval time".
- **Deleted requests are still counted**, with no marker: the pending total can overstate the live workload.
- **The assistant cannot approve**: chasing and approving happen in Spendesk.

## Observed result (demo account, 5 October 2026)

3 pending card requests for EUR 607.00, all created on 9 June 2026, i.e. 118 days waiting: critical band. All three were waiting on the same approver, and none had a cost center. Over 12 months: 11 requests, 8 of them approved (EUR 1,702.98).

## Example prompts
- "Which card requests have been waiting for approval for too long?"
- "Which approver are card requests stuck with?"
- "Are there card requests older than 30 days that I should chase?"